09:10 – 09:55 | Conference Room
More than 80% of AI startups/projects fail not due to technology or talent. In this session, we will focus on 5 key reasons why AI projects have failed and how to avoid them. I will focus on examples from the real-world scenario that caused a major hindrance or roadblock for the success of AI projects during my journey in the past 7 years. These are practical problems that AI entrepreneurs/practitioners do not expect and are not prepared for. This should help anyone in AI avoid these simple but costly mistakes.

09:55 – 10:40 | Conference Room
Emotion analysis is a obvious extension to the popular task of sentiment analysis, however, methods and applications differ. In a first part of this talk, I provide a brief overview on the psychological background on emotions and their purpose and how this leads to challenges when they are estimated from text. I highlight open research questions and possible research directions. The second part is concerned with applications in a variety of areas: What can emotion analysis contribute to the humanities, for instance literary studies? Finally, I will briefly report on use-cases for emotion analysis in a text analysis platform for direct use by customers.

10:40 – 11:00 | Coffee Space

11:00 – 11:30 | Conference Room
This talk will present main goals and results of project ASSISIbf. ASSISIbf aims to establish a robotic society that is able to develop communication channels to animal societies (honeybees & fish swarms) on its own. These robots adapt by evolutionary algorithms until they have learned to interact with animals in a desired way. This new technology is aimed to lay new foundations on the way in which humans can interfere with animal societies. Furthermore, the project results should provide mechanisms that will be used to establish novel paradigm in bio-symbiotic computation.

11:30 -11:50 | Conference Room
Data availability, processing power, and deep learning technique jumpstarted a revolution in computer systems that surpassed humans’ ability in tasks deemed untouchable to machines. State of the art models can translate from one language to another, diagnose cancer, and play games such as chess and poker better than the best human experts. Naturally, we were made to believe that everything was within reach, but that is not always the case. E.g. if you ever had a chance to interact with a chat bot and expected a Star War experience, you were most likely let down by its lack of understanding of even a simplest conversation. In this talk, we will look into several instances of problems today’s AI systems are good at and propose a strategy for building a sustainable and defendable business based on them.

11:50 – 12:30 | Conference Room
The race to build fully autonomous cars has gone into hyperdrive, with major carmakers such as GM, Daimler, BMW and Audi promising SAE Level 5 autonomous driving by sometime in 2021. Can human driving behaviour be replaced by autonomous driving and how? This talk address these challenges. It points out which technologies can be used. It addresses the intelligent and autonomous decisions in real time and show how a few of these capabilities can look like.

12:30 – 13:00 | Conference Room
A lot of time has passed since the first robot was used for Brain biopsy in 1985, where do we stand with medical robotics in 2018? How many of the utopian science fiction scenarios have already turned into reality? The presentation will highlight different medical applications where robotics is already in standard clinical use as well as some research projects that will become the medicine of the future. The robotic solution will only be a first step in this direction. Once there is a machine in the interface between patient and physician, this can be the entry point for new algorithms based on machine learning and artificial intelligence.

13:00 -13:50 | Lobby

13:50 – 14:20 | Conference Room
Transfer learning may be the today the most popular technique for creating deep learning applications. Whenever a deep learning model already exists within a certain domain, transfer learning can be applied. One does not train deep learning models from scratch. Rather, an existing model can be used to transfer knowledge from that model to a new model. For example, if one needs to classify various picture of shoes, an already trained model for classifying pictures of cars, houses and fruit can be used. As the knowledge is being transferred, the training of a new model (for shoes) becomes faster if using an existing model (for cars) than if the training is made from scratch. Using the theory of practopoiesis it is possible to rise transfer learning to a whole new level. Transfer of knowledge can become much more effective. Learning becomes even faster. Fewer examples are needed. Better generalizations is achieved.

14:20 – 14:40 | Conference Room
Implementation of machine and deep-learning in web applications and ERP of one of the largest manufacturers of liquid cooling systems for PCs and workstations – with the purpose of providing near real-time sales forecasts and sentiment analysis enabling rapid response to online user feedback on the market covering more than 30 countries.

14:40 -15:00 | Conference Room
AI-enabling technologies are part of digital transformation initiatives in many organizations. Robotic process automation (RPA) is a catalyst for AI and is fueled by data (from social networks, digital platforms and IoT sensors). The corporate landscape will be changed by „digital workforce“ in automating tasks with RPA. „Digital workers“ (bots) will replace administrative positions through learning and workflow automation and then allow knowledge workers to focus on addressing exceptions or solving problems. GRC areas are typical for our RPA applications in business agility. Our other expertise and digital platforms are for government inspection systems (e-Inspector digital platform). We will present our experience in many countries (with 37 knowledge bases in SE Europe) and especially Agile Inspector app.

15:00 – 15:30 | Conference Room
Modern Deep Learning is where the state of the art research in Machine Learning meets the real-life product development. Numerous breakthroughs have enabled new applications in medical imaging, manufacturing, retail, and every other possible industry. In this talk, we would like to have an overview of the business cases that require deep learning, and have a quick overview of the products available in Azure that allow you to build, train, and deploy, and re-use deep learning models.

15:30 – 15:50 | Conference Room
Cantab PI, a company with offices in Cambridge (UK), Boston and Zagreb, is developing an innovative approach of providing Machine Learning and Artificial Intelligence as a service to companies, with several Fortune 500 companies already clients. Our approach enables implementing a ML/AI layer for bespoke use cases within the existing IT architecture.

We will present functionality of our modules, and will then demonstrate three use cases: Pricing for pharmaceutical products in USA (a client comment: “The best approach in the world. Full stop.”); Behavioural credit scoring (a recommendation from a top global expert: “Ground-breaking and extremely relevant work”); and Dynamical pricing and capacity management in hospitality (a client comment: “It works very, very, very, very, very well”).

15:50 – 16:10 | Conference Room
Google Open Images dataset is the largest of its kind, with more varied and complex bounding-box annotations spanning 500 classes. Google used the dataset to organize the Open Images challenge on Kaggle, the famous online hub for hosting Data Science competitions, with a total reward of 30,000$. Our task was to build the best performing algorithm for automatic object detection, pushing the field of machine vision even further. We present our solution which encompasses state-of-the-art deep learning algorithms, ranking our team 66th out of 454 highly competitive teams from throughout the world.

16:10 – 17:00 | Conference Room (and Around)
The objective of this Panel talk is to sensitize and encourage participants to think forward and to try to imagine what human profiles will we require in the near future. Also, how will we understand technology and be able to apply it and embed it into business. Particularly, this challenge will be addressed in the case of the use of AI technology and the ethical component here present. So, are we ready for all these changes?
List of Panelists:
1. Draženko Kopljar
2. Davor Runje
3. Jan Šnajder
Presenter: Ivana Conjar
Moderator: Ernest Vlačić

16:30 – 00:00 | Conference Venue

09:00 – 09:45 | Conference Room
As Artificial Intelligence (AI) systems are increasingly making decisions that directly affect users and society, many questions raise across social, economic, political, technological, legal, ethical and philosophical issues. Can machines make moral decisions? Should artificial systems ever be treated as ethical entities? What are the legal and ethical consequences of human enhancement technologies, or cyber-genetic technologies? How should moral, societal and legal values be part of the design process? In this talk, we look at ways to ensure ethical behaviour by artificial systems. Given that ethics are dependent on the socio-cultural context and are often only implicit in deliberation processes, methodologies are needed to elicit the values held by designers and stakeholders, and to make these explicit leading to better understanding and trust on artificial autonomous systems. We will in particular focus on the ART principles for AI: Accountability, Responsibility, Transparency.

09:45 – 10:15 | Conference Room
AI is the future – agreed. But is it the future for women? Will women in technology continue to be Alexas and Cortanas? Research proves there are fewer women in technology as they contribute more towards communal goals and problem-solving rather than intellectual challenges. This combined with the threat of being replaceable by AI, is one among the few reasons why women refuse to embrace technology as a career goal. With machine learning and AI taking precedence in the future, this interactive session focuses on Statistics and Emotional Intelligence as a gateway for more women to embrace technology.

10:15 – 11:05 | Conference Room
Development of the AI industry is unstoppable and the implications on our daily life are inevitable. For this reason, there is a great interest for business, academic and public communities to engage in time in a new social, business and technological revolution. But how to go from interest to business? Where to direct resources? What is the best approach for investment in AI? We have gathered people with experience and insight to help us find some answers. Join us in this very interesting conversation.

11:05 – 11:20 | Coffee space

11:25 – 11:45 | Conference Room
The number of farmers who use smartphones is increasing and along with RGB, thermal cameras are becoming more and more available either as smartphone gadgets or as integrated parts of the smartphone. Currently, yield prediction by counting ears and extrapolating the values for the whole field requires ears to be counted manually on the field, and for larger fields samples must be taken from more than one location, which additionally slows down the process and has influence on the accuracy of yield prediction. Here we present the first results of field phenotyping experiments regarding ear segmentation in wheat based on RGB and thermal imaging.

11:45 – 12:15 | Conference Room
The talk will describe recent cutting edge machine learning methods that produce both accurate and understandable models. The methods, based on predictive clustering, can handle different kinds of complex data mining problems, including multi-target prediction, semi-supervised learning and learning from data streams. The talk will also describe applications of these methods in different areas, ranging from sustainable agriculture, through the discovery of new drugs to the optimization of the operation of spacecraft, such as the Mars Express orbiter.

12:15 – 12:35 | Conference Room
Development of supervised convolutional models has resulted in tremendous advances of visual recognition. Today we are able to design systems which classify images, detect faces and recognize people at a near human level. However, challenges such as cross-dataset generalization or adversarial examples indicate that further work is required to reach truly intelligent artificial vision. In this talk I will present current state of visual recognition: operational principles, current performance, main challenges, as well as research directions which promise to push the limits even further.

12:35 – 12:55 | Conference Room
What if your post about a recipe or a game the other night tells more about you than you ever wanted? Maybe even more than your best friends know about you? The field of personality prediction form social media texts has has a lot of traction lately, as it has become obvious that machine learning-based personality prediction models could take personalization of products, services, and marketing to a whole new level. In this talk we’ll review the main directions in the field, and then present our model based on publicly available comments from Reddit, whose accuracy is on par to people taking online questionnaires. We’ll share our insights on future developments in the field, and also briefly touch on the outstanding ethical issues.

12:55 – 13:15 | Coffee space

13:15 – 13:35 | Conference Room
In the talk we’ll look at two commercial systems developed by leveraging research done at the Artificial Intelligence Laboratory at the Jožef Stefan Institute in Ljubljana. The first one is the Event Registry, a global media monitoring tool that aggregates and analyses news content from over 30,000 news sources published globally in 35 languages. By processing this content with a semantic extraction pipeline and linking it across languages, the platform supports powerful queries to inspect media trends in real time. The second application is a monitoring platform for complex systems from the manufacturing, logistics and finance industry. By processing the trail of data produced by the millions of interconnected elements within such systems, potential issues are caught before people would spot them and appropriate reactions are identified.

13:35 – 13:55 | Conference Room
The process of business to business (B2B) sales forecasting is a complex decision-making process, frequently done judgmentally. It can be modeled as a classification problem in machine learning (ML), building upon sales history to predict an outcome for new sales opportunities. However, top performing ML models are frequently black-boxes and do not support transparent reasoning. The presentation will demonstrate an approach to support the decision-maker with explanation of ML predictions and enable interactive evaluation of decision options. In this way, the approach develops into source of new insights, challenges existing beliefs and allows deeper understanding of drivers behind the successful sales.

13:55 – 14:15 | Conference Room
Electronic financial transactions created wide accepted models for payment of goods and services. The expansion of technologies and new applications increased complexity of financial transactions processing environment, and provided further opportunities for already well organized criminal groups. Fraudulent card transactions volumes in EU are in constant increase since year 2011 forcing financial institutions to research and invest into new fraud detection and prevention technologies. Complexity of real-time card fraud detection lies in the poor amount of contextual information and established time limits for processing, classifying and transmitting card authorization response. Erste Group Card Processor is developing in-house AI for real-time fraud detection, running it against existing standard industry real-time fraud detection solution.

14:15 – 14:35 | Conference Room
There are standard ways of increasing revenues in a hospitality industry like increasing capacity, occupancy or a price. There is an alternative, it comes in a shape of a Machine Learning Ensemble and it is called the Amenity Recommender. It’s wrapped in an A.I. platform as a combination of GANs (Generative Adversarial Networks), embedded vectors, missing values imputation techniques and finally, the models itself for predicting guests’ interests.

14:35 – 14:45 | Conference Room
Closing words: we hope conference was great and looking forward to see you at our regular meetups and other conferences. Learn how you can be part of the community!